Reversible De-Identification for lossless image compression using Reversible Watermarking

Abstract

-De-Identification is a process which can be used to ensure privacy by concealing the identity of individuals captured by video surveillance systems. One important challenge is to make the obfuscation process reversible so that the original image/video can be recovered by persons in possession of the right security credentials. This work presents a novel Reversible De-Identification method that can be used in conjunction with any obfuscation process. The residual information needed to reverse the obfuscation process is compressed, authenticated, encrypted and embedded within the obfuscated image using a two-level Reversible Watermarking scheme. The proposed method ensures an overall single-pass embedding capacity of 1.25 bpp, where 99.8% of the images considered required less than 0.8 bpp while none of them required more than 1.1 bpp. Experimental results further demonstrate that the proposed method managed to recover and authenticate all images considered.

k-Same-Net: k-Anonymity with Generative Deep Neural Networks for Face Deidentification

“…Farrugia et al [ 32 ] presents a reversible deidentification approach for losslessly compressed images, which is compatible with formal privacy protection schemes. Reversible deidentification in this case allows for the recovery of the original content containing private information.…”